echidna 0.8.2

A high-performance automatic differentiation library for Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150

#![cfg(feature = "bytecode")]

use echidna::record_multi;
use num_traits::Float;

#[test]
fn record_multi_basic() {
    // f(x,y) = [x+y, x*y]
    let (tape, values) = record_multi(|v| vec![v[0] + v[1], v[0] * v[1]], &[2.0_f64, 3.0]);

    assert_eq!(values.len(), 2);
    assert!((values[0] - 5.0).abs() < 1e-10);
    assert!((values[1] - 6.0).abs() < 1e-10);
    assert_eq!(tape.num_outputs(), 2);
}

#[test]
fn record_multi_output_values() {
    let (tape, values) = record_multi(|v| vec![v[0] + v[1], v[0] * v[1]], &[2.0_f64, 3.0]);

    let out_vals = tape.output_values();
    assert_eq!(out_vals.len(), 2);
    assert!((out_vals[0] - values[0]).abs() < 1e-15);
    assert!((out_vals[1] - values[1]).abs() < 1e-15);
}

#[test]
fn jacobian_basic() {
    // f(x,y) = [x+y, x*y]
    // J = [[1, 1], [y, x]]
    let (mut tape, _) = record_multi(|v| vec![v[0] + v[1], v[0] * v[1]], &[2.0_f64, 3.0]);
    let jac = tape.jacobian(&[2.0, 3.0]);

    assert_eq!(jac.len(), 2);
    assert_eq!(jac[0].len(), 2);
    // Row 0: [1, 1]
    assert!((jac[0][0] - 1.0).abs() < 1e-10);
    assert!((jac[0][1] - 1.0).abs() < 1e-10);
    // Row 1: [y, x] = [3, 2]
    assert!((jac[1][0] - 3.0).abs() < 1e-10);
    assert!((jac[1][1] - 2.0).abs() < 1e-10);
}

#[test]
fn jacobian_vs_forward_mode() {
    // Cross-validate: tape.jacobian() should match echidna::jacobian()
    fn f_forward(x: &[echidna::Dual<f64>]) -> Vec<echidna::Dual<f64>> {
        vec![x[0].sin() + x[1].cos(), x[0] * x[1].exp()]
    }

    let x = [0.7_f64, 0.3];
    let (_, jac_forward) = echidna::jacobian(f_forward, &x);

    let (mut tape, _) = record_multi(|v| vec![v[0].sin() + v[1].cos(), v[0] * v[1].exp()], &x);
    let jac_tape = tape.jacobian(&x);

    for i in 0..2 {
        for j in 0..2 {
            assert!(
                (jac_forward[i][j] - jac_tape[i][j]).abs() < 1e-10,
                "Jacobian mismatch at ({}, {}): fwd={}, tape={}",
                i,
                j,
                jac_forward[i][j],
                jac_tape[i][j]
            );
        }
    }
}

#[test]
fn reverse_seeded_unit_vectors() {
    // reverse_seeded with unit vector e_i should match reverse(output_i)
    let (mut tape, _) = record_multi(|v| vec![v[0] * v[0] + v[1], v[0] * v[1]], &[2.0_f64, 3.0]);
    tape.forward(&[2.0, 3.0]);

    let grad0 = tape.reverse_seeded(&[1.0, 0.0]);
    let grad1 = tape.reverse_seeded(&[0.0, 1.0]);

    let jac = tape.jacobian(&[2.0, 3.0]);

    for j in 0..2 {
        assert!(
            (grad0[j] - jac[0][j]).abs() < 1e-10,
            "seeded[0] mismatch at {}: {} vs {}",
            j,
            grad0[j],
            jac[0][j]
        );
        assert!(
            (grad1[j] - jac[1][j]).abs() < 1e-10,
            "seeded[1] mismatch at {}: {} vs {}",
            j,
            grad1[j],
            jac[1][j]
        );
    }
}

#[test]
fn vjp_multi_vs_vjp() {
    // Compare vjp_multi with manual weighted sum
    let x = [0.7_f64, 0.3];
    let weights = [0.5, -0.3];

    let (mut tape, _) = record_multi(|v| vec![v[0].sin() + v[1].cos(), v[0] * v[1].exp()], &x);
    let vjp_result = tape.vjp_multi(&x, &weights);

    // Compute manually: J^T * w
    let jac = tape.jacobian(&x);
    for j in 0..2 {
        let expected = weights[0] * jac[0][j] + weights[1] * jac[1][j];
        assert!(
            (vjp_result[j] - expected).abs() < 1e-10,
            "vjp_multi mismatch at {}: {} vs {}",
            j,
            vjp_result[j],
            expected
        );
    }
}

#[test]
fn single_output_backward_compat() {
    // Existing single-output methods should work unchanged
    let (mut tape, val) = echidna::record(|v| v[0] * v[0] + v[1] * v[1], &[3.0_f64, 4.0]);
    assert!((val - 25.0).abs() < 1e-10);
    assert_eq!(tape.num_outputs(), 1);

    let g = tape.gradient(&[3.0, 4.0]);
    assert!((g[0] - 6.0).abs() < 1e-10);
    assert!((g[1] - 8.0).abs() < 1e-10);
}

#[test]
fn jacobian_reeval() {
    // Jacobian at different inputs using the same tape
    let (mut tape, _) = record_multi(|v| vec![v[0] * v[1], v[0] + v[1]], &[1.0_f64, 1.0]);

    let jac1 = tape.jacobian(&[2.0, 3.0]);
    // f1 = x*y, f2 = x+y => J = [[y, x], [1, 1]]
    assert!((jac1[0][0] - 3.0).abs() < 1e-10);
    assert!((jac1[0][1] - 2.0).abs() < 1e-10);
    assert!((jac1[1][0] - 1.0).abs() < 1e-10);
    assert!((jac1[1][1] - 1.0).abs() < 1e-10);

    let jac2 = tape.jacobian(&[5.0, 7.0]);
    assert!((jac2[0][0] - 7.0).abs() < 1e-10);
    assert!((jac2[0][1] - 5.0).abs() < 1e-10);
}